US2026065669A1PendingUtilityA1
Method for extracting urban flood inundation extent based on remote sensing images
Assignee: UNIV NANJING INFORMATION SCIENCE & TECHPriority: Aug 27, 2024Filed: Aug 22, 2025Published: Mar 5, 2026
Est. expiryAug 27, 2044(~18.1 yrs left)· nominal 20-yr term from priority
G06V 10/82G06V 10/764G06V 20/13G06T 3/4053G06T 2207/20081G06T 2207/20084G06V 20/176G06T 5/20G06T 5/10G01S 13/89G06T 7/11Y02A10/40G06N 3/0464G06V 10/513G06V 10/20
59
PatentIndex Score
0
Cited by
0
References
0
Claims
Abstract
A method for extracting urban flood inundation extent based on remote sensing images includes: acquiring and preprocessing a Fengyun-4 satellite remote sensing image to obtain a preprocessed remote sensing image, and dividing the preprocessed remote sensing image into a thin cloud area and a thick cloud area; removing thin clouds and thick clouds in the preprocessed remote sensing image; extracting a first water body area from the cloud-removed remote sensing image to obtain a first urban inundation extent; acquiring a radar image at a same region and a same time as the remote sensing image, and extracting a second water body area from the radar image to obtain a second urban inundation extent; and training a super-resolution convolutional neural network according to the first urban inundation extent and the second urban inundation extent, and using a trained super-resolution convolutional neural network to obtain a reconstructed higher-resolution urban flood inundation extent.
Claims
exact text as granted — not AI-modifiedWhat is claimed is:
1 . A method for extracting urban flood inundation extent based on remote sensing images, comprising:
acquiring and preprocessing a remote sensing image from a Fengyun-4 satellite to obtain a preprocessed remote sensing image, and dividing the preprocessed remote sensing image by using a convolutional neural network into a thin cloud area and a thick cloud area; removing thin clouds in the thin cloud area by using homomorphic filtering, and removing thick clouds in the thick cloud area by using a replacement method, to thereby obtain a cloud-removed remote sensing image; extracting a first water body area from the cloud-removed remote sensing image to obtain a first urban inundation extent; acquiring a radar image at a same region and a same time, and extracting a second water body area from the radar image by using a maximum between-class variance method to obtain a second urban inundation extent; and training a super-resolution convolutional neural network according to the first urban inundation extent and the second urban inundation extent to obtain a trained super-resolution convolutional neural network, and using the trained super-resolution convolutional neural network to obtain a reconstructed higher-resolution urban flood inundation extent; wherein the extracting a first water body area from the cloud-removed remote sensing image to obtain a first urban inundation extent comprises:
calculating, based on cloud-removed remote sensing image data, a modified normalized difference water index (MNDWI) according to a green-band reflectance and a red-band reflectance, through the following formula:
MNDWI
=
ρ
G
r
e
e
n
-
ρ
MIR
ρ
G
r
e
e
n
+
ρ
MIR
where ρ Green represents the green-band reflectance, and ρ MIR represents a mid-infrared band reflectance;
classifying a part of the cloud-removed remote sensing image with the MNDWI greater than a threshold MNDWI 0 as a first target water body, classifying another part of the cloud-removed remote sensing image with the MNDWI less than or equal to the threshold as a first background, and determining the part of the cloud-removed remote sensing image corresponding to the first target water body as the first urban inundation extent; and
calculating, based on the first target water body, the first water body area using the following formula:
A
w
=
P
w
A
t
otal
P
total
,
where P w represents a number of water pixels in a study region, P total represents a total number of pixels in the study region, A w represents the first water body area in the study region, and A total represents a total image area of the study region;
wherein the extracting a second water body area from the radar image by using a maximum between-class variance method to obtain a second urban inundation extent comprises:
converting the radar image into a grayscale image and performing normalization on the grayscale image to map pixel values of the grayscale image into a range of 0-255;
counting a number of pixels at each gray level in the grayscale image to form a gray-level histogram;
calculating a total number N of pixels in the grayscale histogram through the following formula:
N
=
∑
(
n
i
)
where n i represents a total number of pixels with a gray level i in the grayscale histogram, and i is in a range of 0-255;
calculating a probability of each gray level, through the following formula:
p
i
=
n
i
/
N
dividing the pixels in the grayscale histogram into two classes according to a pixel value threshold t 0 , where a pixel set with pixel values less than t 0 is recorded as a class A, and a pixel set with pixel values greater than or equal to t 0 is recorded as a class B;
calculating a weight p A (t) of the class A and a weight p B (t) of the class B, through the following formulas:
p
A
(
t
)
=
∑
0
≤
i
<
t
0
p
i
,
p
B
(
t
)
=
∑
t
0
≤
i
<
225
p
i
,
calculating a grayscale mean m A (t) of the class A and a grayscale mean m B (t) of the class B through the following formulas:
m
A
(
t
)
=
∑
0
≤
i
<
t
0
(
i
×
p
i
)
P
A
(
t
)
,
m
B
(
t
)
=
∑
t
0
≤
i
<
225
(
i
×
p
i
)
P
B
(
t
)
,
calculating a grayscale mean of the grayscale image through the following formula:
m
G
=
∑
0
≤
i
≤
255
(
i
×
p
i
)
,
calculating a between-class variance between the class A and the class B through the following formula:
σ
2
=
p
A
(
t
)
×
(
m
A
(
t
)
-
m
G
)
2
+
p
B
(
t
)
×
(
m
B
(
t
)
-
m
G
)
2
,
traversing all target pixel value thresholds to from 0 to 255, calculating the between-class variance σ 2 corresponding to each of the target pixel value thresholds, and finding a pixel value threshold t* of the target pixel value thresholds corresponding to a maximum between-class variance σ 2 of the between-class variances σ 2 corresponding to the target pixel value thresholds; and
binarizing the grayscale image using the pixel value threshold t* of the target pixel value thresholds corresponding to the maximum between-class variance σ 2 , comprising: setting pixels with pixel values less than t* to 0 to represent a second background, and setting pixels with pixel values greater than or equal to t* to 255 to represent a second target water body; and determining a part of the radar image corresponding to the second target water body as the second urban inundation extent; and
calculating the second water body area based on the second target water body.
2 . The method for extracting urban flood inundation extent based on the remote sensing images according to claim 1 , wherein the removing thin clouds in the thin cloud area by using homomorphic filtering comprises:
decomposing the thin cloud area of the remote sensing image into a plurality of single-band remote sensing images, and removing the thin clouds from the plurality of single-band remote sensing images separately.
3 . The method for extracting urban flood inundation extent based on the remote sensing images according to claim 2 , the removing the thin clouds from the plurality of single-band remote sensing images separately comprises:
modeling the thin cloud area of the remote sensing image mathematically as:
f
s
(
x
,
y
)
=
f
i
(
x
,
y
)
×
f
r
(
x
,
y
)
,
where a two-dimensional function ƒ s (x,y) represents the thin cloud area of the remote sensing image, where x and y represent position coordinates of each pixel on a two-dimensional remote sensing image; ƒ i (x,y) represents an incident component contained in the remote sensing image as a thin cloud illumination component, and ƒ r (x,y) represents a reflected component contained in the remote sensing image as a ground reflection component;
transforming logarithmically the thin cloud area of the remote sensing image to separate the thin cloud illumination component and the ground reflection component of the remote sensing image to obtain a logarithmic transformation result Inf s (x,y), through the following formula:
Inf
s
(
x
,
y
)
=
Inf
i
(
x
,
y
)
+
Inf
r
(
x
,
y
)
,
performing a Fourier transform on the logarithmic transformation result Inf s (x,y) to reduce the thin cloud illumination component to obtain a Fourier transform result S(u,v), through the following formulas:
F
{
Inf
s
(
x
,
y
)
=
F
{
Inf
i
(
x
,
y
)
}
+
F
{
Inf
r
(
x
,
y
)
,
let
S
(
u
,
v
)
=
F
{
Inf
s
(
x
,
y
)
}
,
I
(
u
,
v
)
=
F
{
Inf
i
(
x
,
y
)
}
,
R
(
u
,
v
)
=
F
{
Inf
r
(
x
,
y
)
}
,
then
obtaining
:
S
(
u
,
v
)
=
I
(
u
,
v
)
+
R
(
u
,
v
)
,
using a high-pass filter to suppress a low-frequency component I(u,v) and extract a high-frequency component R(u,v) of the Fourier transform result S(u,v) to thereby obtain a high-pass filtered result, through the following formulas:
p
(
u
,
v
)
=
S
(
u
,
v
)
×
H
(
u
,
v
)
=
I
(
u
,
v
)
×
H
(
u
,
v
)
+
R
(
u
,
v
)
×
H
(
u
,
v
)
,
where H(u,v) represents a transfer function of the high-pass filter;
performing an inverse Fourier transform on the high-pass filtered result p(u,v) to obtain an inverse Fourier transform result p(x,y), through the following formulas:
p
(
x
,
y
)
=
F
-
1
{
(
u
,
v
)
}
=
F
-
1
{
I
(
u
,
v
)
×
H
(
x
,
y
)
}
+
F
-
1
{
R
(
u
,
v
)
×
H
(
x
,
y
)
}
,
let
i
(
x
,
y
)
=
F
-
1
{
I
(
u
,
v
)
×
H
(
x
,
y
)
}
,
r
(
x
,
y
)
=
F
-
1
{
R
(
u
,
v
)
×
H
(
x
,
y
)
}
,
then
obtaining
p
(
x
,
y
)
=
i
(
x
,
y
)
+
r
(
x
,
y
)
,
and
performing an exponential transformation on the inverse Fourier transform result p(x,y), through the following formulas:
g
(
x
,
y
)
=
exp
{
i
(
x
,
y
)
}
+
exp
{
r
(
x
,
y
)
}
,
let
i
0
(
x
,
y
)
=
exp
{
i
(
x
,
y
)
}
,
r
0
(
x
,
y
)
=
exp
{
r
(
x
,
y
)
}
,
where i 0 (x,y) represents a processed thin cloud illumination component, and r 0 (x,y) represents a processed ground reflection component;
wherein an expression of the remote sensing image after removing the thin clouds by using the homomorphic filtering is:
g
(
x
,
y
)
=
i
0
(
x
,
y
)
+
r
0
(
x
,
y
)
.
4 . The method for extracting urban flood inundation extent based on the remote sensing images according to claim 1 , the removing thick clouds in the thick cloud area by using a replacement method comprises:
obtaining a cloud-free remote sensing image and a cloud-covered remote sensing image at the same region captured by the Fengyun-4 satellite at different times or by different sensors, performing geocoding to ensure that coordinate systems of the cloud-free remote sensing image and the cloud-covered remote sensing image are consistent, and replacing cloud-covered areas in the cloud-covered remote sensing image with corresponding areas from the cloud-free remote sensing image on registered remote sensing images.
5 . The method for extracting urban flood inundation extent based on the remote sensing images according to claim 1 , wherein the training a super-resolution convolutional neural network according to the first urban inundation extent and the second urban inundation extent to obtain a trained super-resolution convolutional neural network comprises:
using the first urban inundation extent as an input of the super-resolution convolutional neural network, and the second urban inundation extent as an output of the super-resolution convolutional neural network, to train the super-resolution convolutional neural network to thereby obtain the trained super-resolution convolutional neural network.
6 . The method for extracting urban flood inundation extent based on the remote sensing images according to claim 5 , the using the trained super-resolution convolutional neural network to obtain a reconstructed higher-resolution urban flood inundation extent comprises:
inputting the remote sensing image into the trained super-resolution convolutional neural network to obtain the reconstructed higher-resolution urban inundation extent.Join the waitlist — get patent alerts
Track US2026065669A1 — get alerts on status changes and closely related new filings.
We store only your email — no account needed. See our privacy policy.